Advanced Controls & Experimental Settings

The Brainstorm Reactor is engineered for maximum Brain-Quietness: 95% of its cognitive hardening operates invisibly in the backend.

For strategy leads, architects, and power users who want fine-grained control over the AI's reasoning structure, the platform exposes specialized controls in the System Config Sheet, divided into two distinct tiers:

1. Fortgeschritten (Advanced Controls): Production-ready, rock-solid cognitive steering mechanisms organized by scientific impact.
2. Experimentell (Experimental Prototypes): Research-grade heuristics and friction tools (disabled by default).

Part I: Advanced Controls (Fortgeschritten)

The Advanced Controls section is accessible via the Sliders icon (SlidersHorizontal) in the System Config Sheet.

1. Decoupled Scratchpad (PH-SRTP) — Highest Inferences Impact

Setting: [x] Entkoppeltes Scratchpad (PH-SRTP)
Default: Off (false)

What It Does

When an AI reasoning model generates hidden thoughts inside <think> scratchpads, forcing strict style, length, or formatting constraints directly into its generation loop induces severe formatting friction (the Style Tax: up to -12.4% accuracy degradation).

Enabling Decoupled Scratchpad provides format-unconstrained Chain-of-Thought (Zero Style-Tax):

• The hidden <think> trajectory explores the solution space with 100% unconstrained cognitive freedom.
• Literary registers, tone guidelines, and formatting rules are applied strictly post-hoc to the visible output stream.

Recommended Usage

• Enable when: Using reasoning models for complex mathematical, financial, or technical strategy problems.
• Disable when: You want fast, lightweight conversational responses without deep reasoning overhead.

2. Output Matrix Modes (Non-Collapsing Reasoning)

Setting: Ausgabe-Modus (Matrix Mode)
Default: Standard

Standard LLMs tend to "over-smooth" contradictory user requirements into generic, compromise-heavy clichés (Sycophantic Mode Collapse). Matrix output modes prevent state collapse by structuring the response into formal analytical matrices.

Mode Comparison

• Quantum Superposition Matrix: Holds conflicting constraints concurrently valid ($T = 1.0$ for feasibility, $F = 1.0$ for structural impossibility under specific boundary conditions). Instead of collapsing into a vague compromise, it presents the explicit conditions under which each claim holds true.
• Bakhtinian Polyphonic Matrix: Deconstructs the problem into distinct, un-merged stakeholder stances (e.g. Disruptor vs. Auditor vs. Architect). It concludes with an explicit Unresolved Strategic Tension Points section for human executive decision-making.

3. Thinking Patterns (Epistemic Registers)

Setting: Denkmuster (Registers)
Default: Standard (or Persona-coupled default)

Modulates the deductive and syntactic reasoning scaffolding of the model:

• Deductive (kantian): Forces a strict causal chain (Premise → Restriction → Conclusion). Ideal for risk audits and logical verification.
• Compressed (compressed): Focuses on maximum token efficiency and dense Markdown structures. Ideal for pricing tables and technical specs.
• Stakeholder (dickensian): Focuses on goal conflicts, unspoken tensions, and role-based perspectives.
• Forensic (daoist): Objective, 3rd-person, detached analysis without filler phrases.
• Telegram (telegram): Extracted entities and DAG-like relationship structures only.

4. Daily Mindset Prompts (Denk-Impuls Täglich)

Setting: [x] Denk-Impuls (Täglich)
Default: Off (false)

Focuses the analysis in Direct Mode using rotating cognitive frameworks — including cognitive biases, system laws, paradoxes, cybernetic principles, and logical fallacies. Active frames can be cycled, applied, or toggled on the fly.

Part II: Experimental Prototypes (Experimentell)

The Experimental section is marked with the Sprout icon (Sprout) in the System Config Sheet. These features introduce active friction or probabilistic heuristics and are strictly disabled by default.

1. SmartFriction Protection (Anti-Cognitive-Surrender)

Setting: [ ] SmartFriction Schutz
Default: Off (false)

What It Does

Based on the Tri-System Theory & Cognitive Surrender research (Shaw & Nave, Wharton 2026, effect size $Cohen's\ h = 0.81$), humans tend to deactivate internal critical scrutiny (System 2) when AI delivers fluent, confident prose.

SmartFriction Protection activates deliberate friction:

• Socratic Impulses: Automatic surfacing of unbacked premises and probing questions.
• Spatial Canvas Warnings: Client-side heuristic checks that flag premature convergence (low CAD score) or unchallenged hypotheses (Advocatus Diaboli).

2. De-Risking Critic Audit (EVPI)

Setting: [ ] De-Risking Critic Audit (EVPI)
Default: Off (false)

What It Does

Appends a structured Expected Value of Perfect Information (EVPI) risk reduction section at the absolute conclusion of the AI's response stream.

---
### 🔬 De-Risking & EVPI Audit (Expected Value of Perfect Information)
- **Primary Epistemic Bottleneck:** [Key unverified assumption]
- **Current Risk Profile:** ~[5–50]% Uncertainty
- **Highest EVPI Action:** [The 1 empirical question or action that reduces uncertainty to <5%]

Distinction: The Critic vs. De-Risking Audit

• The Critic (Badge below response): An asynchronous second AI model acting as an independent auditor.
• De-Risking Audit (Text suffix): Generated by the primary model itself, identifying the single most important empirical question to lower decision risk.

Decision Matrix: Which Setting to Use?